Up & Over Sampling... Continued


I recently emailed Kevin Halverson of Muse Electronics and asked him for a short layman's description of the difference between digital over sampling and upsampling. Kevin designed the Muse Model Nine and two ninety six that are on Stereophile's Class A+ list, and he is one of the top digital designers in the business. His entire response of 08/11/00 follows: "Dear Craig; Over & upsampling are both Sample Rate Conversion processes (SRC). Both accomplish the same basic goal of increasing the sample rate to increase the image rejection. Neither has any inherent advantage over the other, assuming both are done in a synchronous and integer fashion. The basic differentiation is that upsampling is an external process, oversampling is an internal one. In the case of the Model Two Ninety Six or Model Nine, the internal rates are 352.8 kHz for CDs and either 354 or 768 kHz for DVDs. The present trend towards the use of "upsampling" devices is to improve the performance of poorer performing converters (those with inadequate image rejection). In the case of either the Two Ninety Six or the Model Nine, neither would benefit from any additional image rejection as both already have more than is necessary. I hope this will give you a small amount of insight to the process and all the marketing hype being thrown about. Best Regards, Kevin Halverson". Craig.
garfish
I would also like to add a short paragraph from Madrigal Audio's position paper on this subject that I recently got fom Madrigal's Todd Sutherland. This paper will soon be posted in it's entirety at www.madrigal.com. Quote: "So what's the point? The point is that "upconverting" digital audio to higher rates for conversion has been around , and in fact has been THE OVERWHELMINGLY DOMINENT WAY OF DOING THINGS, since somewhere between 1982 and 1984, depending on who you ask. Historically it has been called "oversampling". But whether you call it "oversampling", "upconversion", "upsampling" or anything else, it remains the same. It is a method by which the originally sampled data is converted to a higher rate for the purpose of better-quality conversion to analog." Because I ferreted out and posted these meaningful responses on this subject, do not consider me an expert-- I'm not. The conclusions of these responses agree well with what Greysquirrel got from Jeff Kalt of Resolution Audio and posted a few days ago. And sometime ago, a Theta Digital Rep. emailed me that their digital engineers considered oversampling and upsampling to be to be same thing. These industry leaders comments/positions have cleared up a lot of confusion about this subject in my mind. Cheers. Craig
Craig, thanks for post, hate to keep beating this horse, but both methods acheive same result, but the "PROCESS" is different, as KH from Muse above states. Some people claim the different process results in different sound.......the August issue of HFNRR has good article on this comparing Wadia to DCS, explaining the two processes, they claim to hear difference and prefer up-sampling used by DCS
Sam...I agree with your post. Each of the manufactures implements this process differently, so they are going to sound different. BTW is HFNRR available on line? Cheers. Craig
It is no surprise that different CD players made by different manufacturers are going to sound different. What made the DCS machine in this case sound better to this particular listener may have nothing to do with upsampling versus oversampling. The analog output stages of these machines are one strong alternate source of sonic differences. I for one don't really know how they could differentiate the sound of upsampling from all the other factors that might affect the sound.
The drift I get from articles in the major press is that "upsampling" improves digital sound by allowing the use of gentler filters when converting to analog. Seems the cruddy sound of "redbook" digital results from the radical "brickwall" filtering used to remove anything over 20Khz.
The audio stages do make a difference. The new MSB Platinum DAC has NO audio output stage in balanced mode and can drive my monoblocks directly. (It also rate converts and has selectable filters.)The lack of an audio stage is part of the reason that the Platinum sounds better than my old Theta Pro Gen Va upgraded to 24/96. However, the professional digital firms like dCs, Z-Systems and SigTech also pride themselves on the implementation of subtly different algorithms. That is also part of why the upgraded Theta sounded better than the non upgraded Theta. When I got a chance to borrow the dCs multi rate converter and the matching DAC, my Theta sounded better with the dCs converting to 24 bit 96. Even just the Z-systems set to output 24 bit dithered made the Theta sounded better. I've played around with a fiar number of combinations in the digital chain and have concluded that the additional processing power available now due to Moore's Law allows for significant improvement of the sound from a Red Book CD. However, I don't think that there is presently a simple specs driven explanation which means you have to listen to the specific piece of gear. Anyone who gets a chance should listen to the dCs, MSB Platinum, and other advanced digital domain devices to hear the very noticable improvements.
Bernstem, I had the same opinion about the HFNRR article. They basically compared two different CD players and concluded they liked one better than the other. Not much different than if we all compared two different CD players and found the one with the pink flamingo painted on top sounded better (more natural). Was it due to the pink flamingo? I say that not to inflame anyone, only to make the point that there are so many "levers" to adjust the sound of a player. In the hands of great designers like Kevin Halverson and Jeff Kalt, great sounding components can be made. Increase the budget (parts quality, R&D etc) and even better sounding components will follow.
Megasam, I think the only difference Kevin Halverson pointed out is that the term "upsampling" has typically been used when speaking about a separate box placed between the transport and processor. That "upsampling" box is simply an oversampling digital filter, basically the same device located internally between the input receiver and the digital-to-analog convertor in a digital processor.
I don't want to sound ignorant about this because I feel I have a fair idea, conceptually, of how the digital process works. Let me quickly say that Garfish and others who spent their time researching this should be commended. Thank you for your efforts and sharing of information. While reading Mr. Halverson's response, images of Audio Alchemy's DTI-PRO32, Genesis Digital Lens, Camelot, etc. flashed through my head. Don't these devices act as the upsampling device? "Neither has any inherent advantage over the other, assuming both are done in a synchronous and integer fashion. The basic differentiation is that upsampling is an external process, oversampling is an internal one." He goes on to say that his product's don't have a need for that (and shouldn't for the $$$!) But when used with "poorer performing converters" increase their performance (to AS good?). Now here's where my ignorance comes in; Would a good D/A converter (under $2K) mixed with an outboard that had a great clock for low jitter, and added dither to the word length (like afforementioned products) be equal to THE BEST D/A converters? Let's also assume that they are in "a synchronous and integer fashion" I assume this is between the "upsampling converter" and the "oversampling converter". Is this like an I2 bus, firewire, digital coax? Or does this mean they need to be on the same digital frequency (i.e. 96KHz,128KHz,etc.)? Now don't anyone jump on me for this, it's just what I came away with from this discussion and Halverson's e-mail. Thanks.
Hi Treyhoss; It seems to me that you may be trying to "unscrute the unscrutable". Each of the components you mention would probably sound different from the others, and you would just have to try it to see if it provided any improvement. And then of course it would be in the context of the rest of your system. I personally think where these outboard upsamplers would be really worth while would be when used with less expensive CD players (as KH noted), but as the quality of the DA conversion process increases, the value of the outboard upsamplers becomes less important-- just my guess. I'm sure much of the technology used to actually build these components is proprietary-- whether they cost a few hundred dollars or thousands like the dCS gear. Cheers. Craig.
Treyhoss, don't forget the analog output section. You could have two digital systems with identical digital conversion processes and parts and they could still sound considerably better/worse/different because of how the analog section is implemented. As for the "synchronous and integer fashion", I'm not exactly sure what is meant by synchronous (though I'm very sure Kevin Halverson does). The integer portion just means that the oversampling should be carried out as an integer multiple of the incoming data rate (eg. 44.1 x 4 = 176.4, 48 x 4 = 192, 44.1 x 8 = 352.8, 44.1 x 16 = 705.6), not a fractional multiple. I'm kind of curious about the synchronous part. Anyone got an answer? Live the good life, Jordan
Craig, many accolades are due you for getting down, and doing some dirty work. Of all the threads flying about this site, this is definitely one of the most educational I've read. Lots of good information here. Great job!
I agree with Trelja! Craig has kept it all together, adding knowledge and understanding through expert testimony along the way. Great job, Craig! And thanks for a copy of the Madrigal paper!! I love the title: "Upconversion and the Emperor's New Clothes".
Hi Greysquirrel, I hear you loud and clear about the analog output section. I guess I was espousing my thoughts about what I was interpreting from Halverson's e-mail. I definately agree that there is much more than number crunching going on, hence the reason you have great players, lousy players and all points between. I guess I was just curious about these upsampling devices that exist out there and if they would bring a good D/A up to par with a great D/A costing much more $$. I certainly acknowledge the performance will only be achieved when/if all of the devices mate well together. -Tony
Trelja-- thanks for the kind words-- just after the truth. And thanks to Greysquirrel for getting Jeff Kalt's thoughts on this thorny issue-- it was an excellent contribution and really started the "hunt". And it would actually be nice to have that post on this thread. It was really Jeff Kalt, Kevin Halverson, and Madrigal's positon paper that provided the "red meat" for us on this issue, so the biggest thanks goes to them for sharing their knowledge. Unscrutable-- whatever it was, I hope it was pleasurable rather than painful:). Cheers. Craig.
Thanks Craig. I'm at work now, but I'll post the Jeff Kalt e-mail when I get off tommorrow morning. Haven't seen the Madrigal paper posted on their site yet. I think it's a great paper to read for anyone interested. Hold tight. Jordan
Here is the e-mail I sent to Jeff Kalt (Resolution Audio)on 7/28/00 concerning upsampling/oversampling and his reply follows: "I've heard good things about the sound of your CD55 CD player. Most people have attributed the sound of this player to the upsampling you employ. I am confused though. Perhaps you can clear something up for me? I asked this question to several other audio enthusiasts, but haven't really gotten a consensus: what are the differences between upsampling and oversampling? Can you explain the difference or are there differences? My understanding is that upsampling and oversampling are basically the same. By upsampling/oversampling the digital filtering can be more aggressive (outside the audio range), leaving only a gentle analog filter before output. Is this basically correct? Upsampling/Oversampling can't actually create information; the process can only allow more accurate retrieval of what is contained in the 16/44.1 signal. Right? As for oversampling, all delta/sigma (1 bit) type DACS must use oversampling? Are the new 24/96 DACS mostly delta/sigma types or are they ladder DACS? The recent attention to upsampling has me wondering if my understanding of the process is correct? Did mfg's just get better at implementing oversampling techniques to get better sound and needed new marketing jargon to draw interest? I appreciate any help you can provide in explaining this somewhat confusing topic. Thank you, Jordan" Jeff Kalts Reply: "Indeed, there is no technical difference between upsampling and oversampling. The only difference I can discern is in the marketing. Indeed, digital filters can be very aggressive above the audio band without the adverse effects that analog brick-wall filters have. This is possible because of FIR (finite-impulse response) filters, which have constant group delay (zero phase effect vs. frequency). There is no physical realization of an FIR filter in analog. Using FIR digital filters allows the analog filter to be relaxed significantly, because the first "images" are located at much higher frequencies. In our cd55, we use a passive third order filter which is down only 0.2 dB at 20 kHz, yet the rejection of the images at 700 kHz is about 60 dB. And indeed, the digital filters do not create information that may have existed before the mic feed was converted to digital. Some external "upsamplers" may by their nature apply some other filter/eq, but this is independent of the a/d - d/a process. You are also correct regarding the delta-sigma dacs. These dacs are rated for maximum input rate, currently as high as 192 kHz. These converters all run at the output at much higher rates -- typically 12 MHz or thereabouts. The better ones from Analog Devices use extra filter stages when the input rate is lower. Essentially, the dacs run, say, 256x at 44.1 or 48 kHz, 128x at 88.2 or 96, and 64x at 176.4 or 192. This puts the noise modulator heart of the converter at the same frequency regardless of input. The best multi-bits, including the PCM1704, run upwards of 800 kHz, which allows 16x at 44.1 (and 8x at 96, and 4x at 192 input rates). In sum, your perception of "market jargon to draw interest" is dead-on. In addition to preying on the consumer base which generally does not have engineering degrees (and some manufacturers as well), these products offer the opportunity to sneak in digital eqs which will absolutely sound different. Better? That's a different story. Finally, we have just started talking to a dealer in Indiana. If all goes well, I'll pass along the info in a couple of days." Regards, Jeff Kalt Resolution Audio resaudio@ix.netcom.com
Jordan; Thanks for posting your Jeff Kalt email, it adds a lot to this thread. Cheers. Craig
Jordan and Craig, thanks for your postings. You guys are great for sharing this information! Jeff Kalt's e-mail really helped "fill in the gaps" and I think I see where I was off in my earlier posting. Thanks again. -Tony
Craig, I saw that Kevin Halverson visited the Resolution CD thread and the "...Pepsi Challenge" thread. That is pretty cool of him to pop in and comment! Did you ask him to visit the site to help explain some of the thorny topics? If so, great job!! Keep livin' the good life! Jordan
Just came across a commentary by Kevin Halverson on the MUSE Web Site in their DOCUMENTS SECTION which I found EXTREMELY interesting and informative entitled -- Sample Rate Converters (up & oversamplers) & their use in Digital Audio CHECK IT OUT AT: http://www.museelectronics.com
I would also suggest reading dCS's white papers on the effects of Jitter and digital "Smearing" at http://www.dcsltd.co.uk/papers.htm Cheers